Word order in ergative languages

[So Haleza Grise]

I remember reading somewhere (Wikipedia? I think. I can't find it in a quick index search of any books on hand) that ergativity is linked to verb-final or verb-initial syntax; ie. that ergative languages universally are not SVO.

The Australian languages I'm familiar with have no fixed word order, so I don't know if the universal applies there. And what about languages with split-ergativity? Does anyone have any insight to this? Or exceptions?

[vec]

We have a discussion ongoing about this in my thread Kiassan turasta word order questions in C&C Quickies.

The situation is confusing and vastly different from language to language, largely depending on degree of ergativity, degree of synthesis and ideosyncracy. But look over that thread, there's a lot if ideas tossed around.

[cromulant]

Ergative systems are found only in SOV and VSO languages. SVO languages are never ergative.

Universal 1267 is absolute.

Wikipedia also makes the claim; here is the reference.

Déchaine and Manfredi (1998) disagree, but admit that "[t]he existence of ergative SVO languages is standardly denied," and go on to list some of the standard deniers.

[vec]

Dyirbal has free word order.

[Drydic]

Dyirbal has free word order.

Dyirbal is required by Australian law to violate a new universal every year.

[con quesa]

Georgian is split-ergative by tense (certain past tense forms trigger ergative alignment, and there's also an odd thing going on with other past/irrealis verb forms where basically the nominative and accusative cases switch), and SVO is definitely a permissible order. Georgian in general has pragmatically-determined word order, and I think prefers SOV, but SVO is definitely grammatical.

[ná'oolkiłí]

Sorry to go off topic, but I must correct con quesa's statement.

While Georgian has a very unusual MSA whose origins can probably be traced to a split-ergative or (split-)active-stative alignment, synchronically the language is basically accusative. There are three patterns of case marking that occur according to verb class and tense.

Code: Select all

   Subj  D.O.  I.O.
A  nom   dat   dat
B  nar   nom   dat
C  dat   nom   (for)

For convenience I've called the two possible objects Direct and Indirect; objects of a Georgian verb cannot always be classified in this way, though.
"Nar" = "narrative case", which is often called "ergative", but I think this name is misleading. (But then again I guess none of the names for the tenses really matter since their use is so idiosyncratic)
Indirect objects in pattern C are marked with the the postposition -თვის -tvis "for". Pattern C also exhibits inversion: the logical subject is marked with an object marker on the verb and the object with subject markers.

What pattern a verb's arguments are marked with is determined by its class (a morphological and general semantic grouping) and tense.

Code: Select all

      I, III  II  IV
Pres    A     A   C
Aor     B     A   C
Perf    C     A   C

"Pres" includes the present, future, imperfective, conditional, subjunctive, and future subjunctive. "Aor" includes the aorist and the optative. "Perf" includes the perfect, the pluperfect, and the perfect subjunctive.
Verb class is very complicated and not necessarily the best way to group verbs other than to show their case marking patterns, but basically they can be summarized thusly: Class I verbs are the most common; they're usually transitive. Class II are usually intransitive, often stative or passive. Class III includes transitive and intransitives. Class IV verbs usually denote feelings or emotions.

To illustrate:

(1) ლევანი წერილს წერს
Levan-i c̣eril-s c̣er-s
Levan-NOM letter-DAT write-3S.SUBJ.PRES
"Levan is writing a letter"

(2) ლევანმა წერილი დაწერა
Levan-ma c̣eril-i da-c̣er-a
Levan-NAR letter-NOM PERFECTIVE-write-3S.SUBJ.AOR
"Levan wrote a letter"

(3) ლევანს წერილი დაუწერია
Levan-s c̣eril-i da-u-c̣er-ia
Levan-DAT letter-NOM PFTV-3S.OBJ-write-3S.SUBJ.PERFECT
"Levan [apparently] has written a letter"

Compare:

(4) ლევანი წავიდა
Levan-i c̣a-vid-a
Levan-NOM PFTV-go-3S.SUBJ.AOR
"Levan went"

If we only look at (2) and (4), Georgian does appear to have ergative case-marking in the aorist and optative. However, go is a Class II verb; it always takes pattern A. When we look at the aorist of a class III verb, we see that the narrative case marks Nominative arguments (subjects transitive and intransitive) rather than Ergative ones (only transitive subjects).

(5) ლევანი ისვენებს
Levan-i isveneb-s
Levan-NOM relax-3S.SUBJ.PRES
"Levan is relaxing"

(6) ლევანმა დაისვენება
Levan-ma da-isveneb-a
Levan-NAR PFTV-relax-3S.SUBJ.AOR
"Levan relaxed"

Instead of being an Ergative case, the narrative is better thought of (in my opinion) as more of a Nominative₂ case (the nominative, of course, being Nominative₁—or Accusative₂).

Some authors maintain Georgian's split ergativity by saying such Class III intransitives are at an underlying level really transitive, and that there is in fact an invisible/underlying/somehow deleted Absolutive argument—ie, at some level (4) is really "Levan relaxed relaxation", or something. This may be the diachronic origin of Pattern B, but I don't find this argument very convincing at all from a synchronic perspective.

Georgian MSA can't easily be neatly classified, but I think of it as a nominative-accusative system that can take three different and independent surface realizations. Instead of having 9 surface cases, however, many of them overlap in shape, thus obscuring this fact. That is to say, we can think of there being the "underlying" (in what sense I can't say; this is more of a explanatory model than an attempt at a "real" explanation) cases Nom₁, Nom₂, Nom₃, Acc₁, etc. However, Nom₁, Acc₂, and Acc₃ are marked with a single surface case (the nominative); Nom₂ with the narrative; and Acc₁, Dat₁, Dat₂, and Nom₃ with the dative.

———

I'm not at an advanced enough level to speak in depth about word order, but I can say con quesa is correct in saying that SOV is default but other orders occur pragmatically.

[cromulant]

Perhaps the universal should be that ergative languages are universally not strictly SVO.

[Drydic]

Probably as good a place as any to ask this. I've been working on a daughterfamily (if you will) of an Ergative language which killed the majority of case marking. I've been told there haven't been any examples found of ergative languages which don't have some form of actual case marking, ie they use strict word order to indicate the ergative/absolutive roles. Is this the case? (my own investigations, albeit limited, haven't found anything yet).

[Xephyr]

Probably as good a place as any to ask this. I've been working on a daughterfamily (if you will) of an Ergative language which killed the majority of case marking. I've been told there haven't been any examples found of ergative languages which don't have some form of actual case marking, ie they use strict word order to indicate the ergative/absolutive roles. Is this the case? (my own investigations, albeit limited, haven't found anything yet).

There's no argument marking on verbs?

[vec]

Probably as good a place as any to ask this. I've been working on a daughterfamily (if you will) of an Ergative language which killed the majority of case marking. I've been told there haven't been any examples found of ergative languages which don't have some form of actual case marking, ie they use strict word order to indicate the ergative/absolutive roles. Is this the case? (my own investigations, albeit limited, haven't found anything yet).

To my knowledge, no language has ergative and absolutive cases that are evenly marked (ie. not distinguished). In all the ones I have found, the ergative is always overtly marked (maybe with a handful of exceptions, but that's always the general rule. No language has the absolutive being more marked morphologically.) What the daughters could develop is using an adposition to mark the ergative; which in turn could become a case suffix in some; remain an adposition in others; or change to nominative – accusative with an ergative flavour, betraying their past.

[Chuma]

First of all, Drydic: Is your parent lang strictly ergative? Since those are very rare, it might be that the situation you describe just hasn't happened, by coincidence. It doesn't sound all that unrealistic.

Second: Suppose it's true that ergative langs are always SOV or VSO. (That's annoying - my conlang is ergative and SVO. Anyway.) And since most are not "pure" ergative, they often have for example the syntactic pivot on the subject. With that in mind, if you remove the overt case marking, what is there to tell you that it's ergative at all? Wouldn't it be indistinguishable from an accusative language? In other words, there could well be ergative languages without overt case marking, we just can't see that they're ergative.

[Vardelm]

Suppose it's true that ergative langs are always SOV or VSO. (That's annoying - my conlang is ergative and SVO. Anyway.)

This makes me think, is there something that absolutely prevents ergative languages from being SVO (or the other word orders)? Or, is it simply that ergativity isn't a common feature and therefore the samples we do have just happen to not be SVO?

I think when creating conlangs we should ask this question about "universals". It's pretty easy to imagine a language that has ergative marking but is SVO. So, why not do it, unless there is that something that makes SVO ergativity impossible.

[cromulant]

No language has the absolutive being more marked morphologically.

Tlapanec.

[vec]

No language has the absolutive being more marked morphologically.

Tlapanec.

I stand corrected. BTW, languages like Tlapanec and Piraha basically prove that us conlangers can do whatever the hell we want.

[Curlyjimsam]

If you have SVO order in an ergative language, then you then have to stipulate what you mean by "S" and "O" - which one, purely in word order terms, includes the argument of intransitive verbs? An SOV ergative language will have NV in intransitive clauses whatever; a VSO one will have VN. But a SVO one could have either order, depending on whether or not the ergativity extends beyond morphology to basic word order.

I've read that languages tend to be nom-acc in terms of word order even if they're ergative - though how you can prove this if SVO doesn't occur I don't know (Chomsky's Extended Projection Principle claims, broadly, that all sentences must fill the subject position. This would, if true, prevent an SVO language with VN in intransitives - i.e. syntactic ergativity). But if it is the case, then SVO order would have preverbal absolutives in intransitives but postverbal ones in transitives, which would perhaps pressure it either to change to morphological nom-acc marking or to switch to SOV (or VSO) word order, which might explain why ergative SVO languages are rare/nonexistent.

(My main conlang is (syntactically) ergative without actually having case marking. This may not be realistic, but it's based on another planet, so my excuse is that its speakers - who are not quite human - have slightly different versions of UG. It's also mostly SOV, so the ergativity is only marginally apparent, e.g. it affects how the relative clause is formed - it's slightly different with absolutive arguments than with ergative ones - and subjunctive clauses (which are SVO) have VN ordering when intransitive. Some dialects are nom-acc, as well.)

[vec]

I only have one erg-abs language, Kiassan turasta. I decided to use topic-comment–based word order in it. The most important argument comes first, followed by the less important one.

[Vardelm]

If you have SVO order in an ergative language, then you then have to stipulate what you mean by "S" and "O"

I would love to explore this (as well as the rest of your post and this thread) in more detail, but have very little time at the moment.

In my (somewhat informed and definitely not formally linguistically educated) perception, it seems that this is the result of trying to impose accusative notions of subject and object onto ergative languages. The more I learn about ergativity, the more it seems to parallel passive constructions in accusative languages. No, I don't want to open up the "ergativity isn't a passive voice" can-of-worms, but from what I've read there have been many who see parallels. I found a paper recently about this, and I'll look for it when I get a chance.

My own conlang is based on that notion. I refer to it as being OVS transitive and SV intransitive. Really, it's PVA transitive. Because it parallels a passive voice from an accusative language (English in particular), you could argue that in a default sentence that the patient is actually the "subject", so is it in actually SVO? I've gone back and forth on this many times in my head. Hopefully I can finish the phonology & morphology sections someday so I can move on to syntax to have a full explanation.

[vec]

In my Encyclopædia Britannica from '62 that used to be my main linguistic source, all ergative languages were referred to as "passive by default, rather than active like English". So someone at least used to agree with that notion.

[cromulant]

The problem is that the terms "subject" and "object" refer bundles of syntactic and semantic properties which have no necessary relation to one another. Syntactically, the "subject" is generally (always?) the one mandatory constituent of a clause, the target of verbal agreement if there is only one, the most easily relativized argument, and the syntactic pivot. Semantically, "subject" is understood to refer to the agent. My understanding of a syntactically ergative language would be one in which the semantically patientive argument had all the syntactic properties that a "subject" does in a nom-acc language, described above. But to call this argument (the absolutive) the "subject" just creates confusion.

[Vardelm]

The problem is that the terms "subject" and "object" refer bundles of syntactic and semantic properties which have no necessary relation to one another.

Good point. For me, this pinpoints where the issues come from.

Syntactically, the "subject" is generally (always?) the one mandatory constituent of a clause, the target of verbal agreement if there is only one, the most easily relativized argument, and the syntactic pivot.

This 1 sentence clarifies for me better than anything I've read as to what constitutes syntax, separate from semantics.

Semantically, "subject" is understood to refer to the agent.

I wonder, is that simply because the vast majority of the world's languages are accusative, and in particular what I would call "semantically accusative"? That is, the subject of the sentence is associated with the agent?

My understanding of a syntactically ergative language would be one in which the semantically patientive argument had all the syntactic properties that a "subject" does in a nom-acc language, described above.

From what I've seen, I would agree. Another good, concise description.

But to call this argument (the absolutive) the "subject" just creates confusion.

I agree, and I would suggest that it's because most languages are accusative in orientation, which makes us associate the subject with the agent. Because most readers will have the same disposition from their L1, flopping these around to more "accurately" describe a syntactically ergative language creates that confusion.

[vec]

Well, strictly, subject has no direct relation to the thematic rule "agent" in most languages. I know laymen will often say that the subject is "the one who does something, the agent" but that's not true. The subject is simply the primary argument of a specific which cannot be altogether omitted – if it is, it is always implied – while other arguments, such as objects, can be omitted without being implied. Usually, in nom-acc languages, only the subject can have the thematic role "agent", but this is not universal among languages, and the subject can take on a lot of other roles (experiencer, undergoer, theme, patient etc.).

[Chuma]

Yes, we discussed that in Vec's word order thread. From what I've heard, most ergative languages still have subjects and objects much like accusative languages. I suppose you could say that's because they're not completely ergative. But "subject" is a vague term anyway sometimes. I agree with Vardelm, that it might be better to refer to them as PVA etc. (I like to say that my conlang has erg-verb-abs order.)

It's also interesting that no one has mentioned the position of the direct vs. indirect object. Actually, it's not the lack of mentioning that's interesting, I'm just saying it's an interesting thing which no one has mentioned... well, you know what I mean.

[Drydic]

I felt like making a list out of this post. LISTY-LIST LIST LIST!

1. There are separate transitive and intransitive personal endings on verbs. I was referring to the lack of extant ergative or absolutive markers (ie case endings).
2. Word order, the parent language is strict as fuck EAV (Ergative-Absolutive-Verb, so functionally equivalent to SOV), but I want to have the daughterlangs switch to EVA (counterpart to SVO, of course) if I can get away with it. And the parent language, as it stands, is strictly ergative. what can I say, I want to make sure I do it right. PURITY OF FORM, PURGE THE SYSTEM, ELIMINATE THE INFECTION. That said, I have inklings that one of the daughterlangs may become Nom-Acc, but it would be taking a ton of substratum and adstratum influence so I could live with that. Maybe.
3. I don't want adpositional marking of the ergative. I'll make the languages Nom-Acc first (which I don't want to do either.) They're originally postpositional, the daughterlangs, but they keep a minimalist group of caseforms, Direct (Erg & Abs), Oblique (adpositional basically), Instrumental, and Dative in pronouns (nouns have a dative postposition taking the Oblique at the moment). Though as was brought up (at least that's how I read it), I might have to have the Dative work differently.
4. see point 2 for why it's erg-abs even without case marking.
5. I haven't thought much of how this system relates to agent/patient, so I'll need to think about that when I'm supposed to be doing something more important.*

*This, of course, is when I get the majority and best of my conlanging work done.
:blows into a horn:
SOUND OFF!

LIST!